Fifth-Order Compact Gas-Kinetic Scheme

• CGKS-5th is a high-order finite-volume scheme that uses compact reconstruction of cell averages and gradient moments to align the numerical domain with the physical flow.
• It employs a BGK-based gas evolution model to generate time-dependent fluxes and interface states, reducing dissipation and enhancing shock resolution.
• Validated in turbulence, shocks, acoustics, and wall-bounded flows, CGKS-5th demonstrates superior computational efficiency compared to lower-order methods.

Searching arXiv for papers on compact gas-kinetic schemes and CGKS-5th to ground the article in the relevant literature. The fifth-order compact gas-kinetic scheme, commonly abbreviated CGKS-5th, denotes a class of high-order finite-volume gas-kinetic schemes in which the BGK model is used to construct time-dependent interface fluxes and time-evolving interface states, while fifth-order spatial accuracy is obtained from compact reconstructions based on cell averages together with gradient-type moments. In the literature, CGKS-5th appears in structured and unstructured forms, in explicit and implicit steady-state solvers, and in extensions for implicit large eddy simulation, shock-acoustic interaction, moving meshes, and wall-modeled turbulent flow. A recurring feature is that both the conservative variables and their gradients are updated from the same gas distribution function, so compactness is tied not only to stencil width but also to the physical domain of dependence embodied by the gas evolution model (Zhao et al., 2022, Zhao et al., 2020, Yang et al., 2023).

1. Emergence and scope within high-order gas-kinetic methods

CGKS-5th belongs to the broader development of high-order gas-kinetic schemes in which time-accurate flux evolution is combined with compact multi-moment reconstruction. Earlier work on high-order GKS emphasized that the gas-kinetic framework naturally supplies flux time derivatives, enabling multi-stage multi-derivative formulations that achieve high temporal accuracy with fewer stages than Runge-Kutta methods; specifically, fifth-order GKS with two and three stages was developed in the MSMD framework (Ji et al., 2017). Compact higher-order GKS with spectral-like resolution then established the central idea that the interface evolution provides both fluxes and interface flow variables, allowing cell averages and cell-averaged gradients to be updated within each time step without introducing separate evolution equations for slopes (Zhao et al., 2019, Zhao et al., 2019).

Within this trajectory, the fifth-order compact variant became a distinct object of study in several settings. For structured turbulence simulation, a fifth-order compact reconstruction in the normal direction was assessed for implicit large eddy simulation and compared with higher-order non-compact reconstructions (Zhao et al., 2022). For three-dimensional unstructured meshes, implicit compact HGKS with Hermite WENO reconstruction and GMRES acceleration was developed for steady compressible flows (Yang et al., 2023). Later work evaluated CGKS-5th against a conventional second-order GKS in scale-resolving compressible simulations and used it as the outer solver in a wall-modeled framework for high-Reynolds-number separated turbulent flows (Yang et al., 27 Aug 2025, Yang et al., 29 Jun 2026).

Representative paper	Setting	Defining feature
(Zhao et al., 2022)	Structured turbulence ILES	5th-order compact normal reconstruction
(Yang et al., 2023)	3D unstructured steady flows	Implicit HWENO + GMRES + CUDA
(Yang et al., 27 Aug 2025)	Compressible scale-resolving simulations	Cost/accuracy comparison with GKS-2nd

This body of work suggests that “CGKS-5th” is best understood as a methodological family rather than a single stencil formula.

2. Kinetic foundation and conservative update

The mathematical foundation is the BGK model for the gas distribution function,

$$\frac{\partial f}{\partial t} + u\frac{\partial f}{\partial x} + v\frac{\partial f}{\partial y} + w\frac{\partial f}{\partial z} = \frac{g-f}{\tau},$$

or, in vector notation,

$$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$

where $f$ is the gas distribution function, $g$ the Maxwellian equilibrium, and $\tau$ the collision time (Zhao et al., 2022, Yang et al., 2023, Yang et al., 29 Jun 2026). Macroscopic conservative variables and fluxes are obtained as moments of $f$, and the finite-volume update is written in semi-discrete form as

$$\frac{d\boldsymbol{Q}_{ijk}}{dt}=\mathcal{L}(\boldsymbol{Q}_{ijk}),$$

or, on unstructured meshes,

$|\Omega_i|\frac{dQ_i}{dt}=\mathcal{L}(Q_i),$

with interface fluxes evaluated by integrating the gas-kinetic solution at Gaussian quadrature points (Zhao et al., 2022, Yang et al., 2023).

A defining property of compact GKS is that the same time-dependent interface solution supplies both the numerical flux and the time-evolving interface conservative variables. As a result, the cell-averaged flow variables and their gradients can be updated from the same physical solution at the interface, rather than from separate governing equations for auxiliary degrees of freedom (Zhao et al., 2019, Zhao et al., 2019). On structured meshes this appears through a Newton-Leibniz update such as

$$(Q_x)_i=\frac{1}{\Delta x}\left(Q_{i+1/2}-Q_{i-1/2}\right),$$

where the interface values are equilibrium face-averaged conservative variables (Zhao et al., 2022). On unstructured meshes the same role is played by Gauss-theorem or Gauss-Green relations, for example

$$|\Omega_k|(\nabla Q)_k=\int_{\partial\Omega_k}Q(t)\,\boldsymbol{\tau}\,dS,$$

with face values taken from moments of the time-evolved gas distribution function (Yang et al., 2023, Zhao et al., 2020).

The significance of this construction is not merely algebraic. In the compact GKS literature, compactness is explicitly linked to the consistency between high-order initial reconstruction and high-order gas evolution, so that the numerical domain of dependence is aligned with the underlying physical one (Zhao et al., 2020, Zhao et al., 2022).

3. Fifth-order compact reconstruction and the meaning of compactness

In the structured formulation assessed for turbulence simulation, the fifth-order compact GKS uses compact HWENO reconstruction in the normal direction and fifth-order linear polynomial reconstruction in the tangential directions (Zhao et al., 2022). The normal-direction reconstruction proceeds in three steps: first, interface conservative variables are converted into cell-averaged derivatives by the Newton-Leibniz formula; second, fifth-order HWENO reconstructs interface states from cell averages and cell-averaged derivatives; third, fifth-order linear reconstruction is applied in tangential directions to obtain values and derivatives at Gaussian quadrature points (Zhao et al., 2022). For the interface $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$0, the compact stencil uses averages from $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$1 and derivatives from $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$2 and $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$3, and the paper contrasts this with non-compact WENO-type schemes that use five cell-averaged values (Zhao et al., 2022).

This structured CGKS-5th is “compact” in a strict stencil sense: for fifth order, it uses only three cell-averaged values and two cell-averaged derivatives. The paper further states that the compact reconstruction has the consistent physical and numerical domains of dependence without employing additional information from cells which have no any direct physical connection with the targeted cell (Zhao et al., 2022). That formulation is central to the compactness claim.

On unstructured meshes, the concrete stencil geometry changes but the principle remains the same. In the implicit three-dimensional compact HGKS, the HWENO reconstruction stencil contains only one level of neighboring cells,

$$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$4

whereas the non-compact WENO version uses two levels of neighbors (Yang et al., 2023). In triangular unstructured CGKS for acoustic and shock wave computation, compactness is defined by the physical domain of dependence and typically corresponds to a nine-cell stencil consisting of the target triangle, its neighbors, and the neighbors’ neighbors adjacent to the central cell’s faces (Zhao et al., 2020). In later structured three-dimensional CGKS-5th formulations, the compact stencil is described in terms of the target cell together with face neighbors and edge neighbors, and a quartic polynomial $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$5 is determined from cell averages, gradients, and directional derivatives by constrained least squares, with GENO used for shock robustness (Yang et al., 27 Aug 2025, Yang et al., 29 Jun 2026).

These variants show that fifth-order compactness in GKS is multi-moment rather than purely minimal-width. The conserved averages are not the only data; gradients and line-averaged or directional derivatives are part of the reconstruction state.

4. Temporal discretization, equilibrium relaxation, and nonlinear stabilization

Many CGKS-5th implementations combine fifth-order spatial reconstruction with a two-stage fourth-order time discretization. This S2O4 strategy is used in the turbulence-oriented structured scheme, in compact unstructured formulations that update averages and gradients from the gas-kinetic interface solution, and in later structured CGKS-5th performance studies (Zhao et al., 2022, Zhao et al., 2020, Yang et al., 27 Aug 2025). The common appeal is efficiency: high temporal accuracy is achieved with two stages and with direct use of time-derivative information supplied by the gas-kinetic flux.

At the same time, the broader GKS literature shows that fifth-order time-accurate gas-kinetic methods also exist. A family of fifth-order schemes with two and three stages was developed using MSMD time stepping, specifically the S3O5 and S2O5 formulations, with 5th-order WENO-Z spatial reconstruction in the tests (Ji et al., 2017). This matters when interpreting the label “CGKS-5th”: in a substantial part of the compact literature, “fifth-order” refers to spatial order, while the time integrator may remain fourth order.

Shock treatment is handled through the kinetic evolution together with nonlinear reconstruction. One formulation emphasizes that the gas evolution model unifies nonlinear and linear reconstructions through a relaxation process from non-equilibrium to equilibrium, with contributions weighted by $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$6 and $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$7; in smooth regions the linear compact reconstruction dominates, while in shock regions the nonlinear part provides the needed dissipation (Zhao et al., 2019). On unstructured meshes, nonlinear reconstruction has been built as a combination of WENO and ENO methodology, using a discontinuous initial reconstruction for the non-equilibrium state and an evolved smooth reconstruction for the equilibrium state (Zhao et al., 2020). Other fifth-order compact implementations use GENO-type switching to lower-order ENO-type sub-stencils near discontinuities (Yang et al., 27 Aug 2025, Yang et al., 29 Jun 2026).

A later very high-order extension replaced classical smoothness indicators with a discontinuity feedback factor, allowing adaptive stencil extension and order reduction while keeping first-order robustness near discontinuities. In that framework, the fifth-order case uses a compact three-cell stencil with cell averages and slopes (Mu et al., 19 Aug 2025). Although this paper is not the canonical CGKS-5th formulation, it clarifies how compact gas-kinetic ideas have been generalized beyond fifth order.

5. Validation in turbulence, shocks, acoustics, and wall-bounded flow

The structured fifth-order compact GKS for ILES was validated first by three-dimensional advection of a density perturbation,

$$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$8

for which CGKS-5th achieved designed fifth-order convergence in $$f_t + \boldsymbol{u}\cdot\nabla f = \frac{g-f}{\tau},$$9, $f$0, and $f$1 norms (Zhao et al., 2022). In the three-dimensional Taylor-Green vortex problem, the compact scheme denoted C5T5 was compared with DNS and with non-compact reconstructions. The reported statistics included the time history of kinetic energy, dissipation rate, enstrophy, and $f$2-criterion isosurfaces, and the paper states that CGKS-5th matches DNS much better than 5th-order non-compact N5T5, exhibits reduced numerical dissipation, and resolves smaller-scale turbulent structures more faithfully (Zhao et al., 2022).

For turbulent channel flow at $f$3 and $f$4, the same study reports that CGKS-5th is much closer to DNS in the mean velocity profile, especially near the centerline; gives more accurate Reynolds stresses and RMS quantities; resolves a higher wavenumber range in the energy spectra; and captures more vortices in $f$5-criterion visualizations (Zhao et al., 2022). The paper further gives explicit friction-Reynolds-number comparisons: at $f$6, CGKS-5th yields $f$7, compared with N5T5 at $f$8 and DNS at $f$9; at $g$0, CGKS-5th gives $g$1, which the paper describes as much closer to DNS $g$2 than lower-order schemes (Zhao et al., 2022).

The compact GKS literature is not limited to nearly incompressible turbulence. Compact higher-order GKS with spectral-like resolution was developed for compressible flows containing both acoustic and shock waves, with the stated property that, with the same stencil of a second-order scheme, a reconstruction up to 8th-order spacial accuracy can be constructed (Zhao et al., 2019). On unstructured meshes, a compact high-order GKS up to sixth order of accuracy was constructed for acoustic and shock wave computation, with validated accuracy, efficiency, and robustness and with the conclusion that beyond the first-order Riemann solver, the use of high-order gas evolution model seems necessary in the development of high-order schemes (Zhao et al., 2020).

Later performance evaluations compared CGKS-5th directly against a conventional second-order gas-kinetic scheme. For subsonic Taylor-Green vortex at $g$3, $g$4, CGKS-5th on a $g$5 mesh was reported as comparable to GKS-2nd on a $g$6 mesh, with computational times of $g$7 versus $g$8 on 2 GPUs, corresponding to an efficiency improvement of about $g$9 (Yang et al., 27 Aug 2025). For supersonic Taylor-Green vortex at $\tau$0, $\tau$1, CGKS-5th on a $\tau$2 mesh was reported as approximately equivalent to GKS-2nd on a $\tau$3 mesh, with $\tau$4 versus $\tau$5 on 4 GPUs, or about $\tau$6 improvement (Yang et al., 27 Aug 2025). In transonic and supersonic turbulent jet flows at matched computational cost, CGKS-5th was reported to resolve substantially finer turbulence and shock structures, while the jet potential core prediction was listed as $\tau$7 for CGKS-5th, compared with experiments at about $\tau$8–$\tau$9, high-order LES at $f$0, and GKS-2nd at $f$1 (Yang et al., 27 Aug 2025).

A recent wall-modeled extension uses CGKS-5th as the outer solver and couples it to a pressure-gradient-based non-equilibrium wall model. In separated turbulent flow past a circular cylinder and in a shock-induced separation bubble on the transonic RAE 2822 airfoil, the framework was reported to reproduce the separated flow structures and markedly improve near-wall predictions over its wall-model-free counterpart, most notably the skin-friction coefficient (Yang et al., 29 Jun 2026). For the RAE 2822 case, the near-wall mesh was described as twentyfold coarsened in the wallnormal direction, with comparable coarsening in other directions, and the wall-model coupling was reported to add less than $f$2 runtime overhead in a multi-GPU implementation (Yang et al., 29 Jun 2026).

6. Implicit formulations, unstructured-mesh acceleration, and current interpretive issues

CGKS-5th has also been developed as an implicit and hardware-conscious algorithm. On three-dimensional unstructured meshes, implicit compact HGKS couples HWENO reconstruction with backward-Euler pseudo-time stepping and GMRES solution of the linearized system. The compact stencil contains one level of neighboring cells, gradients are recomputed using fluxes and the gas-kinetic solution at interfaces, Jacobi iteration is used as the GMRES preconditioner, and the method was implemented on GPUs with CUDA (Yang et al., 2023). In the same work, the timing table was summarized as showing nearly $f$3 speedup on GPU for both compact and non-compact versions versus CPU, while the compact stencil was described as especially suitable for future scaling to multiple GPUs (Yang et al., 2023).

For steady-state computation on 3D unstructured mixed-element meshes, an advanced compact GKS was combined with a geometric multigrid strategy. The scheme evolves cell-averaged flow variables and their gradients on the original mesh; coarse meshes use first-order KFVS; and a three-layer V-cycle with an explicit forward-Euler coarse-grid smoother was reported to produce a convergence rate up to ten times faster than standard CGKS, whereas implicit LU-SGS offered limited convergence acceleration (Liu et al., 2024). The paper also states that the explicit multigrid CGKS is highly scalable and effective for large-scale computations (Liu et al., 2024).

Moving-mesh extensions show a different limitation. An efficient and robust compact ALE gas-kinetic scheme for unstructured meshes uses a memory-reduction reconstruction requiring only inversion of a small $f$4 matrix per cell per time step, which was reported to make a $f$5 speedup in ALE computations (Wang et al., 15 Aug 2025). However, that work also states that the current memory-reduction reconstruction can deliver fifth-order spatial accuracy but the overall scheme remains fourth-order unless the temporal discretization is further advanced (Wang et al., 15 Aug 2025). This is one of the clearest indications that “CGKS-5th” should not be read automatically as “fifth order in both space and time” across all implementations.

Two misconceptions are therefore recurrent. The first is that CGKS-5th is simply a compact WENO scheme. The literature instead treats the gas evolution model as the core mechanism: compactness and high-order accuracy come from the consistency between reconstruction and the high-order gas evolution solution, and the interface solution supplies both fluxes and evolved interface variables (Zhao et al., 2020, Zhao et al., 2019). The second is that compactness has a single universal stencil definition. The cited works define compactness through three-cell-plus-derivative stencils on structured grids, one-level neighbor stencils on unstructured polyhedral grids, or physical-domain-of-dependence stencils on triangular meshes (Zhao et al., 2022, Yang et al., 2023, Zhao et al., 2020). A plausible implication is that the unifying concept is local kinetic multi-moment coupling rather than one fixed reconstruction topology.